Digital Twining a Living Lab

Interview with David Gerber – University of Southern California

At the CDFAM NYC 2025, David Gerber, Professor at the University of Southern California, presented the Living Lab Project at Ginsburg Hall, a LEED Platinum-certified building designed as a digital twin for research.

With 3,500 sensors across 252 devices, the system captures real-time data on building performance to support research across architecture, engineering, AI, and health.

Built on the TraxTwin platform and hosted on Azure, the Living Lab provides open access via API and browser tools. In this interview, Gerber outlines the system architecture, research applications, and cultural challenges faced during development.

Could you start by introducing the University of Southern California’s Living Lab Project and what you will be presenting at CDFAM?

The Living Lab started out as a directive from our University Board to ensure our newest building would achieve a LEED Platinum certification and because they saw this as an opportunity to support research in important ways.

What transpired from numerous stakeholder meetings was the idea that the building needed to be a platform for design sciences and for science and engineering in the broadest possible terms. I then took to heart the input from faculty from across the spectrum and decided the building could become an instrument of science. One that not only provided real time data of the physical plant but one that would provide data of its behaviours and responses to the environment, to the systems, and ultimately to the nature of the building being a Living Lab.

Our mission was to build a system that would support researchers and educational programs as inclusively as possible. In the most tangible sense the building allows for building science, architecture, civil engineering, and structural engineering to observe and use real time data for their own research directions.

In a purposeful extension of the intended use, the building provides data for our AI and machine learning researchers, our embodied and physical AI researchers, for our physiological and medical researchers, and can support many disciplines in between.

The Ginsburg Hall Living Lab is a digital twin built to support a global community of research. It is an instrument of science and design enabling the development of novel technologies, methodologies and policies for improving the built environment. It is also a story and exemplar of collaboration across ownership, project management, design, engineering, computer science, general and sub-contracting, sustainability sciences and fundamental innovation.

The Living Lab uses embedded sensors and digital twin technology for real-time monitoring of building performance. Can you describe the architecture of this system and how the different data sources are integrated?

All the data sources (252 devices with 3,500 sensors), that we currently have in the system, are integrated by the Honeywell Building Automation System (BAS). The architecture of our system in explained in this slide. However, the architecture of the Honeywell system is unknown.

That being said, we hope to connect to additional data sources not coming from the BAS such as the energy production from the solar cells or the lighting data from the lighting system. User-defined datasets can also be included using the LivingLab API, such as wifi data, etc.

What software platforms or frameworks are being used to build and manage the digital twin, and how were these chosen for the project?

The LivingLab system is built with the TraxTwin software that can be used directly through a browser, independently by using the USC Living Lab Sensor API, or indirectly via Google Colab Notebooks that uses the API. These systems were chosen by the Trax development team based on meeting the needs of a living lab system that could be used by faculty and students.

About the API, specifically that there is a very standard and simple REST API. Developers would want to use their own tools, code environment, and software. They may want to take a look at the Colab Notebook for the samples, but they wouldn’t use that if they were making their own system using the API.

How is the collected data processed, stored, and made accessible for research purposes across different disciplines?

The collected data is processed to create multiple levels of roll-ups such as hourly, daily, weekly, and annually. The data is stored as blobs on Microsoft Azure. They are made available via the API or through the user interfaces mentioned above i.e. USC Living Lab, Google Colab Notebooks.

1) Architecture & data-source integration

The Living Lab’s data plane is a straightforward ingestion → processing → storage → access pipeline. We acquire ~252 devices / ~3,500 points primarily from the Honeywell BAS (EBI 600) via OPC UA (through Matrikon Tunneller), with REST pulls where subsystems expose them. A containerized Node.js collector polls/streams on configurable intervals (10 min), normalizes timestamps to UTC, and maps vendor point names into a stable key space: project_id / device_id / sensor_id plus metadata. The collector writes newline-delimited JSON “slices” to Azure Blob by day for raw streams. The same path supports non-BAS feeds (e.g., PV, lighting, meter, weather, occupancy aggregates) without changing the downstream contract.

2) Software stack & rationale

The web UI (TraxTwin) is simple JavaScript portal, that sits over a Node.js/Express REST API that exposes sensors, rollups, and metadata with consistent, versioned contracts. Services are containerized and deployed on Azure (App Service/Container Apps), with CI/CD via GitHub Actions. We chose this stack to minimize activation energy for students/faculty (browser + Colab) while keeping the backend simple and cheap to operate (stateless API, blob-first storage, horizontal scale via containers).

3) Processing, storage, and research access

Raw time series land in Blob Storage as JSON at /raw/{project}/{device}/{sensor}/{YYYY}/{MM}/{DD}.json. Scheduled workers (Node.js) compute rollups at 1, 2, 3, 4, 6, 8, 12, and 24-hour intervals using time-aligned buckets and standard aggregates (min/max/mean/median/stddev/count/sum). Outputs go to /rollups/{project}/{sensor}/{interval_hours}/{YYYY}/{MM}.json. Researchers can (a) query the REST API with filters on project/sensor/interval/time window, (b) pull signed URLs for bulk export (CSV on demand), or (c) work directly in Colab via Python.

What challenges have you encountered in ensuring accuracy, security, and long-term usability of the building performance data?

Security is an ongoing process and challenge. This is addressed by having collaborators register with USC. Accuracy is not a problem. However, our data monitoring does reveal that the system is down from time to time. While we have a self-restarting system, on one occasion the local system had to be restarted by USC staff. Too early to tell, about long-term usability of the data, but the purpose of the roll-ups is to manage the level of detail to help maintain usability of the system.

What do you hope to share with, and learn from, the CDFAM community about applying digital twin technology in real-world building research environments?

A major thread of what I would want to convey are the challenges in the development technologically but more so culturally and anthropologically.

We faced issues of copyright, security, privacy, and bureaucracy. The other is the notion of turning our second most important assets into instruments of science and research for AI to analogue research disciplines. People being the first and our physical plant being the second but historically under-valued.

The other is all about digital twins and the hype versus the actual value and opportunities created and the possible extensions into design, into AI ML research, into health and Human Building Interaction…

its also a chance to discuss true digital twins and digital shadows as opposed to just another BIM.

To explore how data capture and data-driven design are transforming fields from architecture and product design to engineering and manufacturing, join David Gerber and other experts at CDFAM in NYC this year. The symposium gathers leaders applying AI, machine learning, and advanced computation across disciplines.